Distributional regression for seasonal data: an application to river flows

Battle Count: 76

The paper is primarily focused on environmental/hydrological risk modeling for insurance applications. While the distributional regression framework and misspecified model inference techniques could theoretically be adapted for financial time series with seasonal patterns, the direct relevance to quantitative trading is minimal. The seasonal modeling approach and GAMLSS framework could inspire similar approaches for commodity price seasonality or energy market modeling, but this is tangential.

The methodology involves: (1) implementing the extended generalized gamma distribution with numerical stability near nu=0, (2) constructing Fourier basis functions and interaction terms, (3) maximizing a misspecified likelihood, (4) computing sandwich estimators with Tukey-Hanning weighting, (5) implementing stepwise model selection with TIC, and (6) handling multiple stations jointly. While the R 'gamlss' package provides some infrastructure, the extended distribution and misspecified inference require custom implementation. The mathematical framework is well-defined but requires careful numerical handling.

The paper provides detailed mathematical formulations, model selection procedures, and parameter specifications. However, no code or repository is explicitly mentioned. The R package 'gamlss' is referenced for implementation, and the extended generalized gamma implementation is attributed to Perreault et al. (2025). Data from Water Survey of Canada (hydrometric stations) is publicly available but specific station IDs are not given. The 31-day Tukey-Hanning bandwidth and stepwise selection procedure are described in detail.